Leveraging Fine-Tuned YOLOv8 with Transfer Learning for Peritoneal Carcinomatosis Detection

Peritoneal Carcinomatosis (PC) is a severe disease where cancer spreads to the peritoneal lining, and patients with this condition have a high risk of mortality. Early detection of this disease is crucial for improving patient survival chances. This study utilizes the YOLOv8 model to detect PC using a dataset of 4,197 laparoscopy images. YOLOv8 was chosen for its capability in fast and accurate object detection. The model training process involves using pretrained weights for fine-tuning to enhance detection performance. Model hyperparameters were optimized, and data augmentation was applied to improve model generalization. Experimental results show that the YOLOv8m model performs the best, with a mean Average Precision (mAP50) of 0.871, precision of 0.893, recall of 0.785, and an F1 score of 0.84. The use of fine-tuned pre-trained weights on the PC dataset shows a significant improvement in results compared to models trained without pre-trained weights.